Collaborative research in Asia calls for caution when using popular large-scale language model-based (LLM) chatbots as part of public health research and response.
In a study published in BMJ, researchers from the Chinese University of Hong Kong, RMIT University in Vietnam, and the National University of Singapore investigated whether LLM bridges or worsens the digital divide in access to accurate health information. I investigated.
Why is it important?
The research team used OpenAI’s widely used GPT-3.5 chatbot to ask questions about atrial fibrillation symptoms in Vietnamese. The chatbot then responded with an answer related to Parkinson’s disease.
“Misunderstandings in symptom detection and disease guidance can have a significant impact on outbreak management,” said Kwok, one of the researchers and an associate professor of public health and primary health at the Faculty of Medicine, University of Hong Kong, China. Mr. Kinion warned.
The researchers identified problems due to language bias in LLM tools, as there is less training in low-resource languages like Vietnamese (or languages with few digital resources available). According to Dr Arthur Tan, a senior lecturer at RMIT Vietnam University, this results in lower quality answers in languages that are less exposed.
“This disparity in LLM accuracy could exacerbate the digital divide, especially since low-resource languages are primarily spoken in low- to middle-income countries,” he added.
NUS Associate Professor Wilson Tam said AI chatbots should be carefully monitored for accuracy and reliability, “especially when prompted and responses are generated in low-resource languages.” I gave advice. “Providing a fair platform to access health information is beneficial, but ensuring the accuracy of this information is essential to prevent the spread of misinformation.”
Researchers will improve LLM translation capabilities for diverse languages, create and share open source language data and tools to promote AI language inclusivity, and address current limitations in LLM-driven medical communication. I suggested that you deal with it.
“Strengthening the LLM is critical to ensure the provision of accurate, culturally and linguistically appropriate health information, especially in regions vulnerable to infectious disease outbreaks.” CU Medicine A/Kuok The professor emphasized.
bigger trends
In another study, the same research team demonstrated the use of ChatGPT in developing disease transmission models to inform infection control strategies. According to them, the LLM tool acts as a co-pilot in quickly building initial transmission models and different model variants, “significantly” reducing the time required to develop such complex models. It is said that he did.
“In the face of a potential outbreak, rapid response is critical, and LLMs can greatly facilitate preliminary analysis and understanding of infection dynamics for novel pathogens. These systems provide instant modeling support. “This enables real-time scenario analysis and facilitates faster and more informed decision-making,” they noted.
Since the widespread adoption of LLM in the healthcare field last year, there has been a growing body of research testing the accuracy and effectiveness of LLM across a variety of applications.
For example, a study last year confirmed that ChatGPT not only improves clinical decision-making but also provides relevant answers for cardiovascular disease prevention.
Although the tool was reported to have passed the highly competitive US Medical Licensing Examination last year, ChatGPT was also found to have failed tests by the American College of Gastroenterology.
In addition to language bias, a Yale University study published this year revealed “startling” findings that prove ChatGPT’s racial bias. The two versions of the chatbot had a statistically significant difference in simplifying radiology reports when the race of the questioner was specified, the researchers said.
